Providers of mapping-related services are continually challenged to accurately map and represent features and objects (e.g., buildings, trees, cars, other structures, etc.) in a mapping database. For example, with respect to representing buildings and other similar objects, service providers generally begin by identifying the building footprint (e.g., a 2-dimensional polygon) from which to build three-dimensional models. However, many automated and manual processes for extracting building footprints from sensor data (e.g., imagery data) or generated from any other means can often result in noisy polygons (e.g., polygons with many edge orientations not present in the original object being modeled). As a result, service providers face significant technical challenges to efficiently reducing building footprint or polygon noise.